Data from mark-recapture studies are used to estimate population rates such as exploitation, survival, and growth. Many of these applications assume negligible tag loss, so tag shedding can be a significant problem. Various tag shedding models have been developed for use with data from double-tagging experiments, including models to estimate constant instantaneous rates, time-dependent rates, and type I and II shedding rates. In this study, we used conditional (on recaptures) multinomial models implemented using the program SURVIV (G.C. White. 1983. J. Wildl. Manage. 47: 716-728) to estimate tag shedding rates of lake trout (Salvelinus namaycush) and explore various potential sources of variation in these rates. We applied the models to data from several long-term double-tagging experiments with Lake Superior lake trout and estimated shedding rates for anchor tags in hatchery-reared and wild fish and for various tag types applied in these experiments. Estimates of annual tag retention rates for lake trout were fairly high (80-90%), but we found evidence (among wild fish only) that retention rates may be significantly lower in the first year due to type I losses. Annual retention rates for some tag types varied between male and female fish, but there was no consistent pattern across years. Our estimates of annual tag retention rates will be used in future studies of survival rates for these fish.